MIDSCA: Towards a Smart Camera architecture of mobile internet devices

Abstract

Each mobile phone with a built-in CMOS sensor can inherently be seen as smart camera itself. Previous work in the research field of smart camera networks intensively focused on custom hardware solutions to optimise the execution of computationally intensive computer vision algorithms. In contrast, we propose an architecture for smart camera networks consisting of low-cost mobile Internet devices (e.g. mobile phones or PDAs). Due to mass production their price decreases steadily, although their processing capacity as well as their networking abilities increases. New standards like LTE promise bandwidth increases as high as 100 Mbps on the downlink, and up to 50 Mbps on the uplink. In this paper, we present MIDSCA, an architecture for self-organising smart camera networks consisting of mobile Internet devices. One component of our architecture uses visual events themselves to implement an algorithm which is able to synchronise mobile devices with the same field of view with the accuracy of one frame. This synchronisation makes way for future mobile surveillance scenarios as multiple cameras are able to establish a common notion of concurrent events. An initial prototype of MIDSCA has been developed and bases upon a Linux operating system. Evaluations show that our algorithm is highly suitable for mobile devices as it is mostly invariant with respect to distinct perspectives and illuminations.